🚀 go-pugleaf

In ch.comp.os.linux loris.renggli@sphys.unil.ch wrote:
> The test was run with no load on the system and only one user.
> I find these results surprisingly poor; so if anyone has an idea

Yes, they are.

What does  hdparm -T -t /dev/sda  say ?

What about creating the fs -b 4096 and re-running the benchmark ?

How much RAM reported by free ?  What does vmstat 5 say when
run as the same time as Bonnie (mostly read, mostly write, many
swapping ?)

What version of the kernel (uname -a) ?  Red Hat 6.2 is not
exactly a kernel version (and I don't use Red Hat).

I get routinely around 17-18 MByte/s with Seagate Chetah and
AIC7xxx. Although Adaptec controllers are not that fast compared
e.g. to Symbios Logic (Tekram: better, cheaper).

Also, I got upto 70 MByte/s in a RAID0 configuration (45 MByte/s
on a fast machine with RAID5).

loris.renggli@sphys.unil.ch wrote:
> 
> Has anybody out there some experience with the Dell PowerEdge 6300
> running Linux (I run RedHat 6.2)?
> 
> Even though it has AIC 7890 U2W SCSI interfaces and IBM Ultrastar
> Ultra160 hard disk drives, I don't seem to be able to get better
> than 3-10 MB per second read/write rates.
> 
 (detailed bonnie results snipped - see orig. posting)

> The test was run with no load on the system and only one user.
> I find these results surprisingly poor; so if anyone has an idea
> or experience on how to improve SCSI performance, please
> let me know!!

It's a pity I didn't save my bonnie results of yesterday - 
using a size of 1,280 MB (2x RAM) on a 9G IBM DNES ($250) 
attached to a dual-channel SYM21002 (Symbios 53c896, $220) 
on a self-assembled box with Abit BP-6 (2xCel.466, $400), 
but doing UP 2.0.36, my results were (IIRC): 
-writing with putc : ~ 5500 kB/s 15 %CPU 
-rewriting :  ~ 8900 kB/s 
-writing intell. : ~ 10000 kB/s 15 %CPU
-reading per char. : ? (don't know any more, sorry) 
-reading / block : ~ 15000 kB/s 
/sec :  89 60%CPU 

Sorry again for only vaguely recalling the numbers - I'm only 
pretty sure about the first 2 measurements and the seeks/sec. 
But I don't find the time to repeat that now (far too late 
again -sigh!). 
AND : that test was not meant to be too serious - as I was 
preparing a CD image on my (IDE) root FS and burning the 
CD-RW (on the other SCSI channel) during the test (*g*).

BTW : 
# hdparm -tT /dev/sdb4
#this is the partition I used for bonnie yesterday
/dev/sdb4:
 Timing buffer-cache reads:   32 MB in  0.36 seconds =88.89 MB/sec
 Timing buffered disk reads:  16 MB in  0.98 seconds =16.33 MB/sec

# hdparm -tT /dev/sdb 
#seems the test uses the faster beginning of the disk
/dev/sdb:
 Timing buffer-cache reads:   32 MB in  0.36 seconds =88.89 MB/sec
 Timing buffered disk reads:  16 MB in  0.83 seconds =19.28 MB/sec

-What's to say about this, Loris ? 
I really AM surprised about your comparative bad results; I 
knew already that at least in Linux, LSI/Symbios is the better 
choice, but at that extent (and even more if you take prices 
into account) - UHHHH ! 
Unfortunately, I can't give you tuning tips, kernel config 
options etc., as I never had Ad~1 for my private boxes 
(only on the NT machines in our company) - maybe you can 
improve things a little bit by increasing the # of queued 
tags (if that is an option for your HA config). But I'm 
afraid a change of HA would be the better tuning.
I just couldn't resist commenting this "disaster"... 
(and hope nobody will regard this posting as arrogant or 
something; this is not what I intended, please...)

Juergen

Marc SCHAEFER wrote:
> 
> In ch.comp.os.linux loris.renggli@sphys.unil.ch wrote:
> > The test was run with no load on the system and only one user.
> > I find these results surprisingly poor; so if anyone has an idea
> 
> Yes, they are.
> 
> What does  hdparm -T -t /dev/sda  say ?

# hdparm -T -t /dev/sdc1

/dev/sdc1:
 Timing buffer-cache reads:   128 MB in  0.94 seconds =136.17 MB/sec
 Timing buffered disk reads:  64 MB in  3.31 seconds = 19.34 MB/sec

> What about creating the fs -b 4096 and re-running the benchmark ?

The blocksize is already 4k.
(ext2 filesystems are by default created with 4k blocksize under
RH 6.2 for 18GB disks).

> How much RAM reported by free ?  What does vmstat 5 say when
> run as the same time as Bonnie (mostly read, mostly write, many
> swapping ?)

There is 256MB or RAM, the system does not swap when
running bonnie.

> What version of the kernel (uname -a) ?  Red Hat 6.2 is not
> exactly a kernel version (and I don't use Red Hat).

kernel is 2.2.14-5.0

> I get routinely around 17-18 MByte/s with Seagate Chetah and
> AIC7xxx. Although Adaptec controllers are not that fast compared
> e.g. to Symbios Logic (Tekram: better, cheaper).
> 
> Also, I got upto 70 MByte/s in a RAID0 configuration (45 MByte/s
> on a fast machine with RAID5).

I am also going to use software RAID, but prior to that I'd like
to make sure I get the best performance (actually, I noticed the
slowness because it took 70 minutes to reconstruct a 18GB disk
in RAID1).
---------------------------------------------------------------------
Dr Loris Renggli,   BSP,   Universite de Lausanne,   CH-1015 Lausanne
IT manager      e-mail: loris.renggli@sphys.unil.ch
tel: +41-21-692 3603                             fax: +41-21-692 3605

In ch.comp.os.linux loris.renggli@sphys.unil.ch wrote:
> slowness because it took 70 minutes to reconstruct a 18GB disk
> in RAID1).

RAID reconstruction on Linux *is* slow.

However, Bonnie should be fast. Even if you use files of 10x memory
size as I usually do.